From writing a speech on behalf of Einstein to writing code, ChatGPT can do almost anything, but is still vulnerable to errors making it far from perfect. This raises questions about the use of the conversational AI in life and death situations, even though it reportedly diagnosed a condition in a dog when veterinarians missed it.
But Harvard computer scientist Isaac Kohane, who is also a physician himself, has said that ChatGPT-4 is better than many doctors he has come across.
Accuracy with speed?
Afer a test drive of the AI with his colleagues, Kohane found that ChatGPT correctly answers medical licensing questions 90 per cent of the time.
Not only did he find ChatGPT-4 better at taking tests than previous versions, but admitted that it even surpassed some licensed doctors.
It also translated discharged information for a Portuguese speaker and simplified medical jargon enough to be understood by sixth graders.
Kohane gave some details about a child from a physical exam to ChatGPT and added ultrasound as well as hormone levels.
The AI delivered results within seconds and correctly diagnosed a once-in-a-lakh condition just as Kohane himself would.
Is ChatGPT always right?
Apart from emulating doctors, ChatGPT can also scan lengthy reports instantly and provide tips to doctors about the correct, compassionate way to talk to patients.
But Kohane is also anxious as much as he is impressed since researchers still can't say for sure if ChatGPT's insight is safe and effective all the time.
The chatbot itself said that its intelligence is limited to patterns spotted in data and doesn't involve intentionality, in response to queries about its capabilities.
ChatGPT has also been seen making up answers and defying instructions, while the AI has a tendency to insist that it is right even after making errors.
